Most advice on this topic is a list of questions to ask on a call. Those are useful and we have published a set of them on each of our agency shortlists. This page is about the part that comes after: how to structure the engagement so that a good agency can do good work and a bad one becomes obvious quickly.
The premise is that you will not reliably identify the right agency from a pitch. Pitches are a selected sample of an agency’s best outcomes presented by their best communicator. A better strategy is to assume you might be wrong and design the arrangement so that being wrong is cheap and quick to discover.
Red flags that are genuinely diagnostic
A guarantee of rankings. Nobody controls the ranking systems. A guarantee is either meaningless, scoped to terms nobody searches, or a sign the agency intends to manufacture the result in a way you would not sanction.
No named practitioner. If you cannot establish who will choose your targets and approve your briefs, you are buying capacity rather than judgement, and you will get the capacity.
Deliverables measured in hours or tasks. "Twenty hours of SEO" tells you nothing about what arrives. Agencies that cannot describe their output as shipped artefacts usually do not ship many.
An AI-specific package with its own markup. Google states plainly that no additional markup or special optimisation is required for its AI features. Charging for an AI schema implementation is charging for a file that does nothing.
Keyword density in the proposal. Research on live generative engines found this performed about 10% worse than making no changes at all. Its presence in a 2026 proposal is a direct signal about how current the thinking is.
Reluctance to name a failed engagement. Every agency has them. One that cannot name one is either very new or managing you.
How to read a case study properly
Three questions turn a case study from decoration into evidence. What was the starting position? A jump from nothing to modest traffic on a new domain is a different achievement from the same jump on an established one. Over what period? Percentage gains without a timeframe are unfalsifiable. And what else changed? Agencies rarely mention the funding round, the product launch or the rebrand that ran concurrently.
Then ask whether you can speak to the client. Not every reference is available and that is legitimate, but an agency that has none available at all across a whole roster is telling you something.
Finally, check whether the result is verifiable from outside. Impressions and clicks come from Search Console and cannot be independently checked, which is exactly why an agency claiming them should be comfortable showing the account on a call rather than a screenshot in a deck.
Structuring the selection so being wrong is cheap
Start with a paid, scoped first piece of work rather than a retainer. An audit, a technical remediation, a content pilot — something finite with a defined output. You learn more about an agency from one delivered project than from three pitches, and the cost of discovering a mismatch is bounded.
Agree the leading indicators before anything starts, in writing, with dates. Indexation movement, first page-one entries, crawl distribution, a first assistant citation — whatever fits the work. The point is to have something falsifiable at week six, because the alternative is six months of mutual optimism.
Settle the attribution model before the first invoice, not during the first awkward review. If your business judges channels on last click, say so immediately, and make the agency tell you how organic will be credited across your cycle. Programmes that were working get cancelled over this more often than over performance.
And write down what you keep. The pages are obvious. The briefs, the intent map, the facet policy, the documented decisions and the workflows are the part that determines whether your next agency or your first in-house hire starts from a system or from archaeology.
Matching agency size to your situation
A large agency buys you capacity, parallel workstreams and cover when someone leaves. It costs you seniority per hour and usually means the people who pitched are not the people who deliver. If your constraint is volume and coordination across channels, this is the right trade.
A small senior team buys you the opposite: the people who assessed your site are the people who work on it, and decisions get made quickly. It costs you capacity and resilience. If your constraint is judgement on a small number of important decisions, this is the right trade.
We are the second kind, which is a real limitation and not a humblebrag. If a procurement process requires three comparable references and a bench of twenty, we will not clear it, and the agencies on our shortlists that can are genuinely better suited to that situation.
Why comparing on price mostly fails
Of the 33 agencies we read across three verticals, nine publish any price at all, and the published monthly floors spread roughly twelvefold for work described in similar language. That spread is mostly scope — whether implementation is included, how senior the people are, how competitive the category — rather than margin.
The practical consequence is that comparing two quotes is comparing two scopes that have been described to you in marketing language. The more useful comparison is: for this money, how many hours of whose time, producing what artefacts, measured how. Two proposals that look the same on price frequently differ by a factor of two on that question.